Hi,
On Thu, Mar 10, 2022 at 02:51:16PM +0100, Luca Ferrari wrote:
> a friend of mine has shown to me a "strange" configuration of its
> physical replication server (13): he has both primary_conninfo and
> primary_slot_name, with replication slots active when queried on the
> master. So far so good, but in the configuration he has also
> restore_command to restore archived WALs from a centralized location.
> Does this make sense?
> Because if the replica cannot connect to the master, it will not start
> at all (so I guess no restore_command will be executed). On the other
> hand if the replica can connect to the primary the WALs will be
> shipped by means of streaming.
> Am I missing something?
We do have max_slot_wal_keep_size, and otherwise users could drop/create the
slot while the standby is shutdown if the situation is really bad, so it seems
sensible to me.